1 // SPDX-License-Identifier: GPL-2.0-only
5 * Copyright (C) 1991, 1992 Linus Torvalds
9 * 'fork.c' contains the help-routines for the 'fork' system call
10 * (see also entry.S and others).
11 * Fork is rather simple, once you get the hang of it, but the memory
12 * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
15 #include <linux/anon_inodes.h>
16 #include <linux/slab.h>
17 #include <linux/sched/autogroup.h>
18 #include <linux/sched/mm.h>
19 #include <linux/sched/coredump.h>
20 #include <linux/sched/user.h>
21 #include <linux/sched/numa_balancing.h>
22 #include <linux/sched/stat.h>
23 #include <linux/sched/task.h>
24 #include <linux/sched/task_stack.h>
25 #include <linux/sched/cputime.h>
26 #include <linux/seq_file.h>
27 #include <linux/rtmutex.h>
28 #include <linux/init.h>
29 #include <linux/unistd.h>
30 #include <linux/module.h>
31 #include <linux/vmalloc.h>
32 #include <linux/completion.h>
33 #include <linux/personality.h>
34 #include <linux/mempolicy.h>
35 #include <linux/sem.h>
36 #include <linux/file.h>
37 #include <linux/fdtable.h>
38 #include <linux/iocontext.h>
39 #include <linux/key.h>
40 #include <linux/binfmts.h>
41 #include <linux/mman.h>
42 #include <linux/mmu_notifier.h>
45 #include <linux/vmacache.h>
46 #include <linux/nsproxy.h>
47 #include <linux/capability.h>
48 #include <linux/cpu.h>
49 #include <linux/cgroup.h>
50 #include <linux/security.h>
51 #include <linux/hugetlb.h>
52 #include <linux/seccomp.h>
53 #include <linux/swap.h>
54 #include <linux/syscalls.h>
55 #include <linux/jiffies.h>
56 #include <linux/futex.h>
57 #include <linux/compat.h>
58 #include <linux/kthread.h>
59 #include <linux/task_io_accounting_ops.h>
60 #include <linux/rcupdate.h>
61 #include <linux/ptrace.h>
62 #include <linux/mount.h>
63 #include <linux/audit.h>
64 #include <linux/memcontrol.h>
65 #include <linux/ftrace.h>
66 #include <linux/proc_fs.h>
67 #include <linux/profile.h>
68 #include <linux/rmap.h>
69 #include <linux/ksm.h>
70 #include <linux/acct.h>
71 #include <linux/userfaultfd_k.h>
72 #include <linux/tsacct_kern.h>
73 #include <linux/cn_proc.h>
74 #include <linux/freezer.h>
75 #include <linux/delayacct.h>
76 #include <linux/taskstats_kern.h>
77 #include <linux/random.h>
78 #include <linux/tty.h>
79 #include <linux/blkdev.h>
80 #include <linux/fs_struct.h>
81 #include <linux/magic.h>
82 #include <linux/perf_event.h>
83 #include <linux/posix-timers.h>
84 #include <linux/user-return-notifier.h>
85 #include <linux/oom.h>
86 #include <linux/khugepaged.h>
87 #include <linux/signalfd.h>
88 #include <linux/uprobes.h>
89 #include <linux/aio.h>
90 #include <linux/compiler.h>
91 #include <linux/sysctl.h>
92 #include <linux/kcov.h>
93 #include <linux/livepatch.h>
94 #include <linux/thread_info.h>
95 #include <linux/stackleak.h>
96 #include <linux/kasan.h>
97 #include <linux/scs.h>
99 #include <asm/pgalloc.h>
100 #include <linux/uaccess.h>
101 #include <asm/mmu_context.h>
102 #include <asm/cacheflush.h>
103 #include <asm/tlbflush.h>
105 #include <trace/events/sched.h>
107 #define CREATE_TRACE_POINTS
108 #include <trace/events/task.h>
111 * Minimum number of threads to boot the kernel
113 #define MIN_THREADS 20
116 * Maximum number of threads
118 #define MAX_THREADS FUTEX_TID_MASK
121 * Protected counters by write_lock_irq(&tasklist_lock)
123 unsigned long total_forks
; /* Handle normal Linux uptimes. */
124 int nr_threads
; /* The idle threads do not count.. */
126 static int max_threads
; /* tunable limit on nr_threads */
128 #define NAMED_ARRAY_INDEX(x) [x] = __stringify(x)
130 static const char * const resident_page_types
[] = {
131 NAMED_ARRAY_INDEX(MM_FILEPAGES
),
132 NAMED_ARRAY_INDEX(MM_ANONPAGES
),
133 NAMED_ARRAY_INDEX(MM_SWAPENTS
),
134 NAMED_ARRAY_INDEX(MM_SHMEMPAGES
),
137 DEFINE_PER_CPU(unsigned long, process_counts
) = 0;
139 __cacheline_aligned
DEFINE_RWLOCK(tasklist_lock
); /* outer */
141 #ifdef CONFIG_PROVE_RCU
142 int lockdep_tasklist_lock_is_held(void)
144 return lockdep_is_held(&tasklist_lock
);
146 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held
);
147 #endif /* #ifdef CONFIG_PROVE_RCU */
149 int nr_processes(void)
154 for_each_possible_cpu(cpu
)
155 total
+= per_cpu(process_counts
, cpu
);
160 void __weak
arch_release_task_struct(struct task_struct
*tsk
)
164 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
165 static struct kmem_cache
*task_struct_cachep
;
167 static inline struct task_struct
*alloc_task_struct_node(int node
)
169 return kmem_cache_alloc_node(task_struct_cachep
, GFP_KERNEL
, node
);
172 static inline void free_task_struct(struct task_struct
*tsk
)
174 kmem_cache_free(task_struct_cachep
, tsk
);
178 #ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR
181 * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
182 * kmemcache based allocator.
184 # if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
186 #ifdef CONFIG_VMAP_STACK
188 * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
189 * flush. Try to minimize the number of calls by caching stacks.
191 #define NR_CACHED_STACKS 2
192 static DEFINE_PER_CPU(struct vm_struct
*, cached_stacks
[NR_CACHED_STACKS
]);
194 static int free_vm_stack_cache(unsigned int cpu
)
196 struct vm_struct
**cached_vm_stacks
= per_cpu_ptr(cached_stacks
, cpu
);
199 for (i
= 0; i
< NR_CACHED_STACKS
; i
++) {
200 struct vm_struct
*vm_stack
= cached_vm_stacks
[i
];
205 vfree(vm_stack
->addr
);
206 cached_vm_stacks
[i
] = NULL
;
213 static unsigned long *alloc_thread_stack_node(struct task_struct
*tsk
, int node
)
215 #ifdef CONFIG_VMAP_STACK
219 for (i
= 0; i
< NR_CACHED_STACKS
; i
++) {
222 s
= this_cpu_xchg(cached_stacks
[i
], NULL
);
227 /* Clear the KASAN shadow of the stack. */
228 kasan_unpoison_shadow(s
->addr
, THREAD_SIZE
);
230 /* Clear stale pointers from reused stack. */
231 memset(s
->addr
, 0, THREAD_SIZE
);
233 tsk
->stack_vm_area
= s
;
234 tsk
->stack
= s
->addr
;
239 * Allocated stacks are cached and later reused by new threads,
240 * so memcg accounting is performed manually on assigning/releasing
241 * stacks to tasks. Drop __GFP_ACCOUNT.
243 stack
= __vmalloc_node_range(THREAD_SIZE
, THREAD_ALIGN
,
244 VMALLOC_START
, VMALLOC_END
,
245 THREADINFO_GFP
& ~__GFP_ACCOUNT
,
247 0, node
, __builtin_return_address(0));
250 * We can't call find_vm_area() in interrupt context, and
251 * free_thread_stack() can be called in interrupt context,
252 * so cache the vm_struct.
255 tsk
->stack_vm_area
= find_vm_area(stack
);
260 struct page
*page
= alloc_pages_node(node
, THREADINFO_GFP
,
264 tsk
->stack
= page_address(page
);
271 static inline void free_thread_stack(struct task_struct
*tsk
)
273 #ifdef CONFIG_VMAP_STACK
274 struct vm_struct
*vm
= task_stack_vm_area(tsk
);
279 for (i
= 0; i
< THREAD_SIZE
/ PAGE_SIZE
; i
++) {
280 mod_memcg_page_state(vm
->pages
[i
],
281 MEMCG_KERNEL_STACK_KB
,
282 -(int)(PAGE_SIZE
/ 1024));
284 memcg_kmem_uncharge_page(vm
->pages
[i
], 0);
287 for (i
= 0; i
< NR_CACHED_STACKS
; i
++) {
288 if (this_cpu_cmpxchg(cached_stacks
[i
],
289 NULL
, tsk
->stack_vm_area
) != NULL
)
295 vfree_atomic(tsk
->stack
);
300 __free_pages(virt_to_page(tsk
->stack
), THREAD_SIZE_ORDER
);
303 static struct kmem_cache
*thread_stack_cache
;
305 static unsigned long *alloc_thread_stack_node(struct task_struct
*tsk
,
308 unsigned long *stack
;
309 stack
= kmem_cache_alloc_node(thread_stack_cache
, THREADINFO_GFP
, node
);
314 static void free_thread_stack(struct task_struct
*tsk
)
316 kmem_cache_free(thread_stack_cache
, tsk
->stack
);
319 void thread_stack_cache_init(void)
321 thread_stack_cache
= kmem_cache_create_usercopy("thread_stack",
322 THREAD_SIZE
, THREAD_SIZE
, 0, 0,
324 BUG_ON(thread_stack_cache
== NULL
);
329 /* SLAB cache for signal_struct structures (tsk->signal) */
330 static struct kmem_cache
*signal_cachep
;
332 /* SLAB cache for sighand_struct structures (tsk->sighand) */
333 struct kmem_cache
*sighand_cachep
;
335 /* SLAB cache for files_struct structures (tsk->files) */
336 struct kmem_cache
*files_cachep
;
338 /* SLAB cache for fs_struct structures (tsk->fs) */
339 struct kmem_cache
*fs_cachep
;
341 /* SLAB cache for vm_area_struct structures */
342 static struct kmem_cache
*vm_area_cachep
;
344 /* SLAB cache for mm_struct structures (tsk->mm) */
345 static struct kmem_cache
*mm_cachep
;
347 struct vm_area_struct
*vm_area_alloc(struct mm_struct
*mm
)
349 struct vm_area_struct
*vma
;
351 vma
= kmem_cache_alloc(vm_area_cachep
, GFP_KERNEL
);
357 struct vm_area_struct
*vm_area_dup(struct vm_area_struct
*orig
)
359 struct vm_area_struct
*new = kmem_cache_alloc(vm_area_cachep
, GFP_KERNEL
);
363 INIT_LIST_HEAD(&new->anon_vma_chain
);
364 new->vm_next
= new->vm_prev
= NULL
;
369 void vm_area_free(struct vm_area_struct
*vma
)
371 kmem_cache_free(vm_area_cachep
, vma
);
374 static void account_kernel_stack(struct task_struct
*tsk
, int account
)
376 void *stack
= task_stack_page(tsk
);
377 struct vm_struct
*vm
= task_stack_vm_area(tsk
);
379 BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK
) && PAGE_SIZE
% 1024 != 0);
384 BUG_ON(vm
->nr_pages
!= THREAD_SIZE
/ PAGE_SIZE
);
386 for (i
= 0; i
< THREAD_SIZE
/ PAGE_SIZE
; i
++) {
387 mod_zone_page_state(page_zone(vm
->pages
[i
]),
389 PAGE_SIZE
/ 1024 * account
);
393 * All stack pages are in the same zone and belong to the
396 struct page
*first_page
= virt_to_page(stack
);
398 mod_zone_page_state(page_zone(first_page
), NR_KERNEL_STACK_KB
,
399 THREAD_SIZE
/ 1024 * account
);
401 mod_memcg_obj_state(stack
, MEMCG_KERNEL_STACK_KB
,
402 account
* (THREAD_SIZE
/ 1024));
406 static int memcg_charge_kernel_stack(struct task_struct
*tsk
)
408 #ifdef CONFIG_VMAP_STACK
409 struct vm_struct
*vm
= task_stack_vm_area(tsk
);
415 for (i
= 0; i
< THREAD_SIZE
/ PAGE_SIZE
; i
++) {
417 * If memcg_kmem_charge_page() fails, page->mem_cgroup
418 * pointer is NULL, and both memcg_kmem_uncharge_page()
419 * and mod_memcg_page_state() in free_thread_stack()
420 * will ignore this page. So it's safe.
422 ret
= memcg_kmem_charge_page(vm
->pages
[i
], GFP_KERNEL
,
427 mod_memcg_page_state(vm
->pages
[i
],
428 MEMCG_KERNEL_STACK_KB
,
436 static void release_task_stack(struct task_struct
*tsk
)
438 if (WARN_ON(tsk
->state
!= TASK_DEAD
))
439 return; /* Better to leak the stack than to free prematurely */
441 account_kernel_stack(tsk
, -1);
442 free_thread_stack(tsk
);
444 #ifdef CONFIG_VMAP_STACK
445 tsk
->stack_vm_area
= NULL
;
449 #ifdef CONFIG_THREAD_INFO_IN_TASK
450 void put_task_stack(struct task_struct
*tsk
)
452 if (refcount_dec_and_test(&tsk
->stack_refcount
))
453 release_task_stack(tsk
);
457 void free_task(struct task_struct
*tsk
)
461 #ifndef CONFIG_THREAD_INFO_IN_TASK
463 * The task is finally done with both the stack and thread_info,
466 release_task_stack(tsk
);
469 * If the task had a separate stack allocation, it should be gone
472 WARN_ON_ONCE(refcount_read(&tsk
->stack_refcount
) != 0);
474 rt_mutex_debug_task_free(tsk
);
475 ftrace_graph_exit_task(tsk
);
476 put_seccomp_filter(tsk
);
477 arch_release_task_struct(tsk
);
478 if (tsk
->flags
& PF_KTHREAD
)
479 free_kthread_struct(tsk
);
480 free_task_struct(tsk
);
482 EXPORT_SYMBOL(free_task
);
485 static __latent_entropy
int dup_mmap(struct mm_struct
*mm
,
486 struct mm_struct
*oldmm
)
488 struct vm_area_struct
*mpnt
, *tmp
, *prev
, **pprev
;
489 struct rb_node
**rb_link
, *rb_parent
;
491 unsigned long charge
;
494 uprobe_start_dup_mmap();
495 if (mmap_write_lock_killable(oldmm
)) {
497 goto fail_uprobe_end
;
499 flush_cache_dup_mm(oldmm
);
500 uprobe_dup_mmap(oldmm
, mm
);
502 * Not linked in yet - no deadlock potential:
504 mmap_write_lock_nested(mm
, SINGLE_DEPTH_NESTING
);
506 /* No ordering required: file already has been exposed. */
507 RCU_INIT_POINTER(mm
->exe_file
, get_mm_exe_file(oldmm
));
509 mm
->total_vm
= oldmm
->total_vm
;
510 mm
->data_vm
= oldmm
->data_vm
;
511 mm
->exec_vm
= oldmm
->exec_vm
;
512 mm
->stack_vm
= oldmm
->stack_vm
;
514 rb_link
= &mm
->mm_rb
.rb_node
;
517 retval
= ksm_fork(mm
, oldmm
);
520 retval
= khugepaged_fork(mm
, oldmm
);
525 for (mpnt
= oldmm
->mmap
; mpnt
; mpnt
= mpnt
->vm_next
) {
528 if (mpnt
->vm_flags
& VM_DONTCOPY
) {
529 vm_stat_account(mm
, mpnt
->vm_flags
, -vma_pages(mpnt
));
534 * Don't duplicate many vmas if we've been oom-killed (for
537 if (fatal_signal_pending(current
)) {
541 if (mpnt
->vm_flags
& VM_ACCOUNT
) {
542 unsigned long len
= vma_pages(mpnt
);
544 if (security_vm_enough_memory_mm(oldmm
, len
)) /* sic */
548 tmp
= vm_area_dup(mpnt
);
551 retval
= vma_dup_policy(mpnt
, tmp
);
553 goto fail_nomem_policy
;
555 retval
= dup_userfaultfd(tmp
, &uf
);
557 goto fail_nomem_anon_vma_fork
;
558 if (tmp
->vm_flags
& VM_WIPEONFORK
) {
560 * VM_WIPEONFORK gets a clean slate in the child.
561 * Don't prepare anon_vma until fault since we don't
562 * copy page for current vma.
564 tmp
->anon_vma
= NULL
;
565 } else if (anon_vma_fork(tmp
, mpnt
))
566 goto fail_nomem_anon_vma_fork
;
567 tmp
->vm_flags
&= ~(VM_LOCKED
| VM_LOCKONFAULT
);
570 struct inode
*inode
= file_inode(file
);
571 struct address_space
*mapping
= file
->f_mapping
;
574 if (tmp
->vm_flags
& VM_DENYWRITE
)
575 atomic_dec(&inode
->i_writecount
);
576 i_mmap_lock_write(mapping
);
577 if (tmp
->vm_flags
& VM_SHARED
)
578 atomic_inc(&mapping
->i_mmap_writable
);
579 flush_dcache_mmap_lock(mapping
);
580 /* insert tmp into the share list, just after mpnt */
581 vma_interval_tree_insert_after(tmp
, mpnt
,
583 flush_dcache_mmap_unlock(mapping
);
584 i_mmap_unlock_write(mapping
);
588 * Clear hugetlb-related page reserves for children. This only
589 * affects MAP_PRIVATE mappings. Faults generated by the child
590 * are not guaranteed to succeed, even if read-only
592 if (is_vm_hugetlb_page(tmp
))
593 reset_vma_resv_huge_pages(tmp
);
596 * Link in the new vma and copy the page table entries.
599 pprev
= &tmp
->vm_next
;
603 __vma_link_rb(mm
, tmp
, rb_link
, rb_parent
);
604 rb_link
= &tmp
->vm_rb
.rb_right
;
605 rb_parent
= &tmp
->vm_rb
;
608 if (!(tmp
->vm_flags
& VM_WIPEONFORK
))
609 retval
= copy_page_range(mm
, oldmm
, mpnt
);
611 if (tmp
->vm_ops
&& tmp
->vm_ops
->open
)
612 tmp
->vm_ops
->open(tmp
);
617 /* a new mm has just been created */
618 retval
= arch_dup_mmap(oldmm
, mm
);
620 mmap_write_unlock(mm
);
622 mmap_write_unlock(oldmm
);
623 dup_userfaultfd_complete(&uf
);
625 uprobe_end_dup_mmap();
627 fail_nomem_anon_vma_fork
:
628 mpol_put(vma_policy(tmp
));
633 vm_unacct_memory(charge
);
637 static inline int mm_alloc_pgd(struct mm_struct
*mm
)
639 mm
->pgd
= pgd_alloc(mm
);
640 if (unlikely(!mm
->pgd
))
645 static inline void mm_free_pgd(struct mm_struct
*mm
)
647 pgd_free(mm
, mm
->pgd
);
650 static int dup_mmap(struct mm_struct
*mm
, struct mm_struct
*oldmm
)
652 mmap_write_lock(oldmm
);
653 RCU_INIT_POINTER(mm
->exe_file
, get_mm_exe_file(oldmm
));
654 mmap_write_unlock(oldmm
);
657 #define mm_alloc_pgd(mm) (0)
658 #define mm_free_pgd(mm)
659 #endif /* CONFIG_MMU */
661 static void check_mm(struct mm_struct
*mm
)
665 BUILD_BUG_ON_MSG(ARRAY_SIZE(resident_page_types
) != NR_MM_COUNTERS
,
666 "Please make sure 'struct resident_page_types[]' is updated as well");
668 for (i
= 0; i
< NR_MM_COUNTERS
; i
++) {
669 long x
= atomic_long_read(&mm
->rss_stat
.count
[i
]);
672 pr_alert("BUG: Bad rss-counter state mm:%p type:%s val:%ld\n",
673 mm
, resident_page_types
[i
], x
);
676 if (mm_pgtables_bytes(mm
))
677 pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n",
678 mm_pgtables_bytes(mm
));
680 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
681 VM_BUG_ON_MM(mm
->pmd_huge_pte
, mm
);
685 #define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
686 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
689 * Called when the last reference to the mm
690 * is dropped: either by a lazy thread or by
691 * mmput. Free the page directory and the mm.
693 void __mmdrop(struct mm_struct
*mm
)
695 BUG_ON(mm
== &init_mm
);
696 WARN_ON_ONCE(mm
== current
->mm
);
697 WARN_ON_ONCE(mm
== current
->active_mm
);
700 mmu_notifier_subscriptions_destroy(mm
);
702 put_user_ns(mm
->user_ns
);
705 EXPORT_SYMBOL_GPL(__mmdrop
);
707 static void mmdrop_async_fn(struct work_struct
*work
)
709 struct mm_struct
*mm
;
711 mm
= container_of(work
, struct mm_struct
, async_put_work
);
715 static void mmdrop_async(struct mm_struct
*mm
)
717 if (unlikely(atomic_dec_and_test(&mm
->mm_count
))) {
718 INIT_WORK(&mm
->async_put_work
, mmdrop_async_fn
);
719 schedule_work(&mm
->async_put_work
);
723 static inline void free_signal_struct(struct signal_struct
*sig
)
725 taskstats_tgid_free(sig
);
726 sched_autogroup_exit(sig
);
728 * __mmdrop is not safe to call from softirq context on x86 due to
729 * pgd_dtor so postpone it to the async context
732 mmdrop_async(sig
->oom_mm
);
733 kmem_cache_free(signal_cachep
, sig
);
736 static inline void put_signal_struct(struct signal_struct
*sig
)
738 if (refcount_dec_and_test(&sig
->sigcnt
))
739 free_signal_struct(sig
);
742 void __put_task_struct(struct task_struct
*tsk
)
744 WARN_ON(!tsk
->exit_state
);
745 WARN_ON(refcount_read(&tsk
->usage
));
746 WARN_ON(tsk
== current
);
749 task_numa_free(tsk
, true);
750 security_task_free(tsk
);
752 delayacct_tsk_free(tsk
);
753 put_signal_struct(tsk
->signal
);
755 if (!profile_handoff_task(tsk
))
758 EXPORT_SYMBOL_GPL(__put_task_struct
);
760 void __init __weak
arch_task_cache_init(void) { }
765 static void set_max_threads(unsigned int max_threads_suggested
)
768 unsigned long nr_pages
= totalram_pages();
771 * The number of threads shall be limited such that the thread
772 * structures may only consume a small part of the available memory.
774 if (fls64(nr_pages
) + fls64(PAGE_SIZE
) > 64)
775 threads
= MAX_THREADS
;
777 threads
= div64_u64((u64
) nr_pages
* (u64
) PAGE_SIZE
,
778 (u64
) THREAD_SIZE
* 8UL);
780 if (threads
> max_threads_suggested
)
781 threads
= max_threads_suggested
;
783 max_threads
= clamp_t(u64
, threads
, MIN_THREADS
, MAX_THREADS
);
786 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
787 /* Initialized by the architecture: */
788 int arch_task_struct_size __read_mostly
;
791 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
792 static void task_struct_whitelist(unsigned long *offset
, unsigned long *size
)
794 /* Fetch thread_struct whitelist for the architecture. */
795 arch_thread_struct_whitelist(offset
, size
);
798 * Handle zero-sized whitelist or empty thread_struct, otherwise
799 * adjust offset to position of thread_struct in task_struct.
801 if (unlikely(*size
== 0))
804 *offset
+= offsetof(struct task_struct
, thread
);
806 #endif /* CONFIG_ARCH_TASK_STRUCT_ALLOCATOR */
808 void __init
fork_init(void)
811 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
812 #ifndef ARCH_MIN_TASKALIGN
813 #define ARCH_MIN_TASKALIGN 0
815 int align
= max_t(int, L1_CACHE_BYTES
, ARCH_MIN_TASKALIGN
);
816 unsigned long useroffset
, usersize
;
818 /* create a slab on which task_structs can be allocated */
819 task_struct_whitelist(&useroffset
, &usersize
);
820 task_struct_cachep
= kmem_cache_create_usercopy("task_struct",
821 arch_task_struct_size
, align
,
822 SLAB_PANIC
|SLAB_ACCOUNT
,
823 useroffset
, usersize
, NULL
);
826 /* do the arch specific task caches init */
827 arch_task_cache_init();
829 set_max_threads(MAX_THREADS
);
831 init_task
.signal
->rlim
[RLIMIT_NPROC
].rlim_cur
= max_threads
/2;
832 init_task
.signal
->rlim
[RLIMIT_NPROC
].rlim_max
= max_threads
/2;
833 init_task
.signal
->rlim
[RLIMIT_SIGPENDING
] =
834 init_task
.signal
->rlim
[RLIMIT_NPROC
];
836 for (i
= 0; i
< UCOUNT_COUNTS
; i
++) {
837 init_user_ns
.ucount_max
[i
] = max_threads
/2;
840 #ifdef CONFIG_VMAP_STACK
841 cpuhp_setup_state(CPUHP_BP_PREPARE_DYN
, "fork:vm_stack_cache",
842 NULL
, free_vm_stack_cache
);
847 lockdep_init_task(&init_task
);
851 int __weak
arch_dup_task_struct(struct task_struct
*dst
,
852 struct task_struct
*src
)
858 void set_task_stack_end_magic(struct task_struct
*tsk
)
860 unsigned long *stackend
;
862 stackend
= end_of_stack(tsk
);
863 *stackend
= STACK_END_MAGIC
; /* for overflow detection */
866 static struct task_struct
*dup_task_struct(struct task_struct
*orig
, int node
)
868 struct task_struct
*tsk
;
869 unsigned long *stack
;
870 struct vm_struct
*stack_vm_area __maybe_unused
;
873 if (node
== NUMA_NO_NODE
)
874 node
= tsk_fork_get_node(orig
);
875 tsk
= alloc_task_struct_node(node
);
879 stack
= alloc_thread_stack_node(tsk
, node
);
883 if (memcg_charge_kernel_stack(tsk
))
886 stack_vm_area
= task_stack_vm_area(tsk
);
888 err
= arch_dup_task_struct(tsk
, orig
);
891 * arch_dup_task_struct() clobbers the stack-related fields. Make
892 * sure they're properly initialized before using any stack-related
896 #ifdef CONFIG_VMAP_STACK
897 tsk
->stack_vm_area
= stack_vm_area
;
899 #ifdef CONFIG_THREAD_INFO_IN_TASK
900 refcount_set(&tsk
->stack_refcount
, 1);
906 err
= scs_prepare(tsk
, node
);
910 #ifdef CONFIG_SECCOMP
912 * We must handle setting up seccomp filters once we're under
913 * the sighand lock in case orig has changed between now and
914 * then. Until then, filter must be NULL to avoid messing up
915 * the usage counts on the error path calling free_task.
917 tsk
->seccomp
.filter
= NULL
;
920 setup_thread_stack(tsk
, orig
);
921 clear_user_return_notifier(tsk
);
922 clear_tsk_need_resched(tsk
);
923 set_task_stack_end_magic(tsk
);
925 #ifdef CONFIG_STACKPROTECTOR
926 tsk
->stack_canary
= get_random_canary();
928 if (orig
->cpus_ptr
== &orig
->cpus_mask
)
929 tsk
->cpus_ptr
= &tsk
->cpus_mask
;
932 * One for the user space visible state that goes away when reaped.
933 * One for the scheduler.
935 refcount_set(&tsk
->rcu_users
, 2);
936 /* One for the rcu users */
937 refcount_set(&tsk
->usage
, 1);
938 #ifdef CONFIG_BLK_DEV_IO_TRACE
941 tsk
->splice_pipe
= NULL
;
942 tsk
->task_frag
.page
= NULL
;
943 tsk
->wake_q
.next
= NULL
;
945 account_kernel_stack(tsk
, 1);
949 #ifdef CONFIG_FAULT_INJECTION
953 #ifdef CONFIG_BLK_CGROUP
954 tsk
->throttle_queue
= NULL
;
955 tsk
->use_memdelay
= 0;
959 tsk
->active_memcg
= NULL
;
964 free_thread_stack(tsk
);
966 free_task_struct(tsk
);
970 __cacheline_aligned_in_smp
DEFINE_SPINLOCK(mmlist_lock
);
972 static unsigned long default_dump_filter
= MMF_DUMP_FILTER_DEFAULT
;
974 static int __init
coredump_filter_setup(char *s
)
976 default_dump_filter
=
977 (simple_strtoul(s
, NULL
, 0) << MMF_DUMP_FILTER_SHIFT
) &
978 MMF_DUMP_FILTER_MASK
;
982 __setup("coredump_filter=", coredump_filter_setup
);
984 #include <linux/init_task.h>
986 static void mm_init_aio(struct mm_struct
*mm
)
989 spin_lock_init(&mm
->ioctx_lock
);
990 mm
->ioctx_table
= NULL
;
994 static __always_inline
void mm_clear_owner(struct mm_struct
*mm
,
995 struct task_struct
*p
)
999 WRITE_ONCE(mm
->owner
, NULL
);
1003 static void mm_init_owner(struct mm_struct
*mm
, struct task_struct
*p
)
1010 static void mm_init_uprobes_state(struct mm_struct
*mm
)
1012 #ifdef CONFIG_UPROBES
1013 mm
->uprobes_state
.xol_area
= NULL
;
1017 static struct mm_struct
*mm_init(struct mm_struct
*mm
, struct task_struct
*p
,
1018 struct user_namespace
*user_ns
)
1021 mm
->mm_rb
= RB_ROOT
;
1022 mm
->vmacache_seqnum
= 0;
1023 atomic_set(&mm
->mm_users
, 1);
1024 atomic_set(&mm
->mm_count
, 1);
1026 INIT_LIST_HEAD(&mm
->mmlist
);
1027 mm
->core_state
= NULL
;
1028 mm_pgtables_bytes_init(mm
);
1031 atomic64_set(&mm
->pinned_vm
, 0);
1032 memset(&mm
->rss_stat
, 0, sizeof(mm
->rss_stat
));
1033 spin_lock_init(&mm
->page_table_lock
);
1034 spin_lock_init(&mm
->arg_lock
);
1035 mm_init_cpumask(mm
);
1037 mm_init_owner(mm
, p
);
1038 RCU_INIT_POINTER(mm
->exe_file
, NULL
);
1039 mmu_notifier_subscriptions_init(mm
);
1040 init_tlb_flush_pending(mm
);
1041 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
1042 mm
->pmd_huge_pte
= NULL
;
1044 mm_init_uprobes_state(mm
);
1047 mm
->flags
= current
->mm
->flags
& MMF_INIT_MASK
;
1048 mm
->def_flags
= current
->mm
->def_flags
& VM_INIT_DEF_MASK
;
1050 mm
->flags
= default_dump_filter
;
1054 if (mm_alloc_pgd(mm
))
1057 if (init_new_context(p
, mm
))
1058 goto fail_nocontext
;
1060 mm
->user_ns
= get_user_ns(user_ns
);
1071 * Allocate and initialize an mm_struct.
1073 struct mm_struct
*mm_alloc(void)
1075 struct mm_struct
*mm
;
1081 memset(mm
, 0, sizeof(*mm
));
1082 return mm_init(mm
, current
, current_user_ns());
1085 static inline void __mmput(struct mm_struct
*mm
)
1087 VM_BUG_ON(atomic_read(&mm
->mm_users
));
1089 uprobe_clear_state(mm
);
1092 khugepaged_exit(mm
); /* must run before exit_mmap */
1094 mm_put_huge_zero_page(mm
);
1095 set_mm_exe_file(mm
, NULL
);
1096 if (!list_empty(&mm
->mmlist
)) {
1097 spin_lock(&mmlist_lock
);
1098 list_del(&mm
->mmlist
);
1099 spin_unlock(&mmlist_lock
);
1102 module_put(mm
->binfmt
->module
);
1107 * Decrement the use count and release all resources for an mm.
1109 void mmput(struct mm_struct
*mm
)
1113 if (atomic_dec_and_test(&mm
->mm_users
))
1116 EXPORT_SYMBOL_GPL(mmput
);
1119 static void mmput_async_fn(struct work_struct
*work
)
1121 struct mm_struct
*mm
= container_of(work
, struct mm_struct
,
1127 void mmput_async(struct mm_struct
*mm
)
1129 if (atomic_dec_and_test(&mm
->mm_users
)) {
1130 INIT_WORK(&mm
->async_put_work
, mmput_async_fn
);
1131 schedule_work(&mm
->async_put_work
);
1137 * set_mm_exe_file - change a reference to the mm's executable file
1139 * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1141 * Main users are mmput() and sys_execve(). Callers prevent concurrent
1142 * invocations: in mmput() nobody alive left, in execve task is single
1143 * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
1144 * mm->exe_file, but does so without using set_mm_exe_file() in order
1145 * to do avoid the need for any locks.
1147 void set_mm_exe_file(struct mm_struct
*mm
, struct file
*new_exe_file
)
1149 struct file
*old_exe_file
;
1152 * It is safe to dereference the exe_file without RCU as
1153 * this function is only called if nobody else can access
1154 * this mm -- see comment above for justification.
1156 old_exe_file
= rcu_dereference_raw(mm
->exe_file
);
1159 get_file(new_exe_file
);
1160 rcu_assign_pointer(mm
->exe_file
, new_exe_file
);
1166 * get_mm_exe_file - acquire a reference to the mm's executable file
1168 * Returns %NULL if mm has no associated executable file.
1169 * User must release file via fput().
1171 struct file
*get_mm_exe_file(struct mm_struct
*mm
)
1173 struct file
*exe_file
;
1176 exe_file
= rcu_dereference(mm
->exe_file
);
1177 if (exe_file
&& !get_file_rcu(exe_file
))
1182 EXPORT_SYMBOL(get_mm_exe_file
);
1185 * get_task_exe_file - acquire a reference to the task's executable file
1187 * Returns %NULL if task's mm (if any) has no associated executable file or
1188 * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1189 * User must release file via fput().
1191 struct file
*get_task_exe_file(struct task_struct
*task
)
1193 struct file
*exe_file
= NULL
;
1194 struct mm_struct
*mm
;
1199 if (!(task
->flags
& PF_KTHREAD
))
1200 exe_file
= get_mm_exe_file(mm
);
1205 EXPORT_SYMBOL(get_task_exe_file
);
1208 * get_task_mm - acquire a reference to the task's mm
1210 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
1211 * this kernel workthread has transiently adopted a user mm with use_mm,
1212 * to do its AIO) is not set and if so returns a reference to it, after
1213 * bumping up the use count. User must release the mm via mmput()
1214 * after use. Typically used by /proc and ptrace.
1216 struct mm_struct
*get_task_mm(struct task_struct
*task
)
1218 struct mm_struct
*mm
;
1223 if (task
->flags
& PF_KTHREAD
)
1231 EXPORT_SYMBOL_GPL(get_task_mm
);
1233 struct mm_struct
*mm_access(struct task_struct
*task
, unsigned int mode
)
1235 struct mm_struct
*mm
;
1238 err
= mutex_lock_killable(&task
->signal
->exec_update_mutex
);
1240 return ERR_PTR(err
);
1242 mm
= get_task_mm(task
);
1243 if (mm
&& mm
!= current
->mm
&&
1244 !ptrace_may_access(task
, mode
)) {
1246 mm
= ERR_PTR(-EACCES
);
1248 mutex_unlock(&task
->signal
->exec_update_mutex
);
1253 static void complete_vfork_done(struct task_struct
*tsk
)
1255 struct completion
*vfork
;
1258 vfork
= tsk
->vfork_done
;
1259 if (likely(vfork
)) {
1260 tsk
->vfork_done
= NULL
;
1266 static int wait_for_vfork_done(struct task_struct
*child
,
1267 struct completion
*vfork
)
1271 freezer_do_not_count();
1272 cgroup_enter_frozen();
1273 killed
= wait_for_completion_killable(vfork
);
1274 cgroup_leave_frozen(false);
1279 child
->vfork_done
= NULL
;
1283 put_task_struct(child
);
1287 /* Please note the differences between mmput and mm_release.
1288 * mmput is called whenever we stop holding onto a mm_struct,
1289 * error success whatever.
1291 * mm_release is called after a mm_struct has been removed
1292 * from the current process.
1294 * This difference is important for error handling, when we
1295 * only half set up a mm_struct for a new process and need to restore
1296 * the old one. Because we mmput the new mm_struct before
1297 * restoring the old one. . .
1298 * Eric Biederman 10 January 1998
1300 static void mm_release(struct task_struct
*tsk
, struct mm_struct
*mm
)
1302 uprobe_free_utask(tsk
);
1304 /* Get rid of any cached register state */
1305 deactivate_mm(tsk
, mm
);
1308 * Signal userspace if we're not exiting with a core dump
1309 * because we want to leave the value intact for debugging
1312 if (tsk
->clear_child_tid
) {
1313 if (!(tsk
->signal
->flags
& SIGNAL_GROUP_COREDUMP
) &&
1314 atomic_read(&mm
->mm_users
) > 1) {
1316 * We don't check the error code - if userspace has
1317 * not set up a proper pointer then tough luck.
1319 put_user(0, tsk
->clear_child_tid
);
1320 do_futex(tsk
->clear_child_tid
, FUTEX_WAKE
,
1321 1, NULL
, NULL
, 0, 0);
1323 tsk
->clear_child_tid
= NULL
;
1327 * All done, finally we can wake up parent and return this mm to him.
1328 * Also kthread_stop() uses this completion for synchronization.
1330 if (tsk
->vfork_done
)
1331 complete_vfork_done(tsk
);
1334 void exit_mm_release(struct task_struct
*tsk
, struct mm_struct
*mm
)
1336 futex_exit_release(tsk
);
1337 mm_release(tsk
, mm
);
1340 void exec_mm_release(struct task_struct
*tsk
, struct mm_struct
*mm
)
1342 futex_exec_release(tsk
);
1343 mm_release(tsk
, mm
);
1347 * dup_mm() - duplicates an existing mm structure
1348 * @tsk: the task_struct with which the new mm will be associated.
1349 * @oldmm: the mm to duplicate.
1351 * Allocates a new mm structure and duplicates the provided @oldmm structure
1354 * Return: the duplicated mm or NULL on failure.
1356 static struct mm_struct
*dup_mm(struct task_struct
*tsk
,
1357 struct mm_struct
*oldmm
)
1359 struct mm_struct
*mm
;
1366 memcpy(mm
, oldmm
, sizeof(*mm
));
1368 if (!mm_init(mm
, tsk
, mm
->user_ns
))
1371 err
= dup_mmap(mm
, oldmm
);
1375 mm
->hiwater_rss
= get_mm_rss(mm
);
1376 mm
->hiwater_vm
= mm
->total_vm
;
1378 if (mm
->binfmt
&& !try_module_get(mm
->binfmt
->module
))
1384 /* don't put binfmt in mmput, we haven't got module yet */
1386 mm_init_owner(mm
, NULL
);
1393 static int copy_mm(unsigned long clone_flags
, struct task_struct
*tsk
)
1395 struct mm_struct
*mm
, *oldmm
;
1398 tsk
->min_flt
= tsk
->maj_flt
= 0;
1399 tsk
->nvcsw
= tsk
->nivcsw
= 0;
1400 #ifdef CONFIG_DETECT_HUNG_TASK
1401 tsk
->last_switch_count
= tsk
->nvcsw
+ tsk
->nivcsw
;
1402 tsk
->last_switch_time
= 0;
1406 tsk
->active_mm
= NULL
;
1409 * Are we cloning a kernel thread?
1411 * We need to steal a active VM for that..
1413 oldmm
= current
->mm
;
1417 /* initialize the new vmacache entries */
1418 vmacache_flush(tsk
);
1420 if (clone_flags
& CLONE_VM
) {
1427 mm
= dup_mm(tsk
, current
->mm
);
1433 tsk
->active_mm
= mm
;
1440 static int copy_fs(unsigned long clone_flags
, struct task_struct
*tsk
)
1442 struct fs_struct
*fs
= current
->fs
;
1443 if (clone_flags
& CLONE_FS
) {
1444 /* tsk->fs is already what we want */
1445 spin_lock(&fs
->lock
);
1447 spin_unlock(&fs
->lock
);
1451 spin_unlock(&fs
->lock
);
1454 tsk
->fs
= copy_fs_struct(fs
);
1460 static int copy_files(unsigned long clone_flags
, struct task_struct
*tsk
)
1462 struct files_struct
*oldf
, *newf
;
1466 * A background process may not have any files ...
1468 oldf
= current
->files
;
1472 if (clone_flags
& CLONE_FILES
) {
1473 atomic_inc(&oldf
->count
);
1477 newf
= dup_fd(oldf
, &error
);
1487 static int copy_io(unsigned long clone_flags
, struct task_struct
*tsk
)
1490 struct io_context
*ioc
= current
->io_context
;
1491 struct io_context
*new_ioc
;
1496 * Share io context with parent, if CLONE_IO is set
1498 if (clone_flags
& CLONE_IO
) {
1500 tsk
->io_context
= ioc
;
1501 } else if (ioprio_valid(ioc
->ioprio
)) {
1502 new_ioc
= get_task_io_context(tsk
, GFP_KERNEL
, NUMA_NO_NODE
);
1503 if (unlikely(!new_ioc
))
1506 new_ioc
->ioprio
= ioc
->ioprio
;
1507 put_io_context(new_ioc
);
1513 static int copy_sighand(unsigned long clone_flags
, struct task_struct
*tsk
)
1515 struct sighand_struct
*sig
;
1517 if (clone_flags
& CLONE_SIGHAND
) {
1518 refcount_inc(¤t
->sighand
->count
);
1521 sig
= kmem_cache_alloc(sighand_cachep
, GFP_KERNEL
);
1522 RCU_INIT_POINTER(tsk
->sighand
, sig
);
1526 refcount_set(&sig
->count
, 1);
1527 spin_lock_irq(¤t
->sighand
->siglock
);
1528 memcpy(sig
->action
, current
->sighand
->action
, sizeof(sig
->action
));
1529 spin_unlock_irq(¤t
->sighand
->siglock
);
1531 /* Reset all signal handler not set to SIG_IGN to SIG_DFL. */
1532 if (clone_flags
& CLONE_CLEAR_SIGHAND
)
1533 flush_signal_handlers(tsk
, 0);
1538 void __cleanup_sighand(struct sighand_struct
*sighand
)
1540 if (refcount_dec_and_test(&sighand
->count
)) {
1541 signalfd_cleanup(sighand
);
1543 * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1544 * without an RCU grace period, see __lock_task_sighand().
1546 kmem_cache_free(sighand_cachep
, sighand
);
1551 * Initialize POSIX timer handling for a thread group.
1553 static void posix_cpu_timers_init_group(struct signal_struct
*sig
)
1555 struct posix_cputimers
*pct
= &sig
->posix_cputimers
;
1556 unsigned long cpu_limit
;
1558 cpu_limit
= READ_ONCE(sig
->rlim
[RLIMIT_CPU
].rlim_cur
);
1559 posix_cputimers_group_init(pct
, cpu_limit
);
1562 static int copy_signal(unsigned long clone_flags
, struct task_struct
*tsk
)
1564 struct signal_struct
*sig
;
1566 if (clone_flags
& CLONE_THREAD
)
1569 sig
= kmem_cache_zalloc(signal_cachep
, GFP_KERNEL
);
1574 sig
->nr_threads
= 1;
1575 atomic_set(&sig
->live
, 1);
1576 refcount_set(&sig
->sigcnt
, 1);
1578 /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1579 sig
->thread_head
= (struct list_head
)LIST_HEAD_INIT(tsk
->thread_node
);
1580 tsk
->thread_node
= (struct list_head
)LIST_HEAD_INIT(sig
->thread_head
);
1582 init_waitqueue_head(&sig
->wait_chldexit
);
1583 sig
->curr_target
= tsk
;
1584 init_sigpending(&sig
->shared_pending
);
1585 INIT_HLIST_HEAD(&sig
->multiprocess
);
1586 seqlock_init(&sig
->stats_lock
);
1587 prev_cputime_init(&sig
->prev_cputime
);
1589 #ifdef CONFIG_POSIX_TIMERS
1590 INIT_LIST_HEAD(&sig
->posix_timers
);
1591 hrtimer_init(&sig
->real_timer
, CLOCK_MONOTONIC
, HRTIMER_MODE_REL
);
1592 sig
->real_timer
.function
= it_real_fn
;
1595 task_lock(current
->group_leader
);
1596 memcpy(sig
->rlim
, current
->signal
->rlim
, sizeof sig
->rlim
);
1597 task_unlock(current
->group_leader
);
1599 posix_cpu_timers_init_group(sig
);
1601 tty_audit_fork(sig
);
1602 sched_autogroup_fork(sig
);
1604 sig
->oom_score_adj
= current
->signal
->oom_score_adj
;
1605 sig
->oom_score_adj_min
= current
->signal
->oom_score_adj_min
;
1607 mutex_init(&sig
->cred_guard_mutex
);
1608 mutex_init(&sig
->exec_update_mutex
);
1613 static void copy_seccomp(struct task_struct
*p
)
1615 #ifdef CONFIG_SECCOMP
1617 * Must be called with sighand->lock held, which is common to
1618 * all threads in the group. Holding cred_guard_mutex is not
1619 * needed because this new task is not yet running and cannot
1622 assert_spin_locked(¤t
->sighand
->siglock
);
1624 /* Ref-count the new filter user, and assign it. */
1625 get_seccomp_filter(current
);
1626 p
->seccomp
= current
->seccomp
;
1629 * Explicitly enable no_new_privs here in case it got set
1630 * between the task_struct being duplicated and holding the
1631 * sighand lock. The seccomp state and nnp must be in sync.
1633 if (task_no_new_privs(current
))
1634 task_set_no_new_privs(p
);
1637 * If the parent gained a seccomp mode after copying thread
1638 * flags and between before we held the sighand lock, we have
1639 * to manually enable the seccomp thread flag here.
1641 if (p
->seccomp
.mode
!= SECCOMP_MODE_DISABLED
)
1642 set_tsk_thread_flag(p
, TIF_SECCOMP
);
1646 SYSCALL_DEFINE1(set_tid_address
, int __user
*, tidptr
)
1648 current
->clear_child_tid
= tidptr
;
1650 return task_pid_vnr(current
);
1653 static void rt_mutex_init_task(struct task_struct
*p
)
1655 raw_spin_lock_init(&p
->pi_lock
);
1656 #ifdef CONFIG_RT_MUTEXES
1657 p
->pi_waiters
= RB_ROOT_CACHED
;
1658 p
->pi_top_task
= NULL
;
1659 p
->pi_blocked_on
= NULL
;
1663 static inline void init_task_pid_links(struct task_struct
*task
)
1667 for (type
= PIDTYPE_PID
; type
< PIDTYPE_MAX
; ++type
) {
1668 INIT_HLIST_NODE(&task
->pid_links
[type
]);
1673 init_task_pid(struct task_struct
*task
, enum pid_type type
, struct pid
*pid
)
1675 if (type
== PIDTYPE_PID
)
1676 task
->thread_pid
= pid
;
1678 task
->signal
->pids
[type
] = pid
;
1681 static inline void rcu_copy_process(struct task_struct
*p
)
1683 #ifdef CONFIG_PREEMPT_RCU
1684 p
->rcu_read_lock_nesting
= 0;
1685 p
->rcu_read_unlock_special
.s
= 0;
1686 p
->rcu_blocked_node
= NULL
;
1687 INIT_LIST_HEAD(&p
->rcu_node_entry
);
1688 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1689 #ifdef CONFIG_TASKS_RCU
1690 p
->rcu_tasks_holdout
= false;
1691 INIT_LIST_HEAD(&p
->rcu_tasks_holdout_list
);
1692 p
->rcu_tasks_idle_cpu
= -1;
1693 #endif /* #ifdef CONFIG_TASKS_RCU */
1694 #ifdef CONFIG_TASKS_TRACE_RCU
1695 p
->trc_reader_nesting
= 0;
1696 p
->trc_reader_special
.s
= 0;
1697 INIT_LIST_HEAD(&p
->trc_holdout_list
);
1698 #endif /* #ifdef CONFIG_TASKS_TRACE_RCU */
1701 struct pid
*pidfd_pid(const struct file
*file
)
1703 if (file
->f_op
== &pidfd_fops
)
1704 return file
->private_data
;
1706 return ERR_PTR(-EBADF
);
1709 static int pidfd_release(struct inode
*inode
, struct file
*file
)
1711 struct pid
*pid
= file
->private_data
;
1713 file
->private_data
= NULL
;
1718 #ifdef CONFIG_PROC_FS
1720 * pidfd_show_fdinfo - print information about a pidfd
1721 * @m: proc fdinfo file
1722 * @f: file referencing a pidfd
1725 * This function will print the pid that a given pidfd refers to in the
1726 * pid namespace of the procfs instance.
1727 * If the pid namespace of the process is not a descendant of the pid
1728 * namespace of the procfs instance 0 will be shown as its pid. This is
1729 * similar to calling getppid() on a process whose parent is outside of
1730 * its pid namespace.
1733 * If pid namespaces are supported then this function will also print
1734 * the pid of a given pidfd refers to for all descendant pid namespaces
1735 * starting from the current pid namespace of the instance, i.e. the
1736 * Pid field and the first entry in the NSpid field will be identical.
1737 * If the pid namespace of the process is not a descendant of the pid
1738 * namespace of the procfs instance 0 will be shown as its first NSpid
1739 * entry and no others will be shown.
1740 * Note that this differs from the Pid and NSpid fields in
1741 * /proc/<pid>/status where Pid and NSpid are always shown relative to
1742 * the pid namespace of the procfs instance. The difference becomes
1743 * obvious when sending around a pidfd between pid namespaces from a
1744 * different branch of the tree, i.e. where no ancestoral relation is
1745 * present between the pid namespaces:
1746 * - create two new pid namespaces ns1 and ns2 in the initial pid
1747 * namespace (also take care to create new mount namespaces in the
1748 * new pid namespace and mount procfs)
1749 * - create a process with a pidfd in ns1
1750 * - send pidfd from ns1 to ns2
1751 * - read /proc/self/fdinfo/<pidfd> and observe that both Pid and NSpid
1752 * have exactly one entry, which is 0
1754 static void pidfd_show_fdinfo(struct seq_file
*m
, struct file
*f
)
1756 struct pid
*pid
= f
->private_data
;
1757 struct pid_namespace
*ns
;
1760 if (likely(pid_has_task(pid
, PIDTYPE_PID
))) {
1761 ns
= proc_pid_ns(file_inode(m
->file
)->i_sb
);
1762 nr
= pid_nr_ns(pid
, ns
);
1765 seq_put_decimal_ll(m
, "Pid:\t", nr
);
1767 #ifdef CONFIG_PID_NS
1768 seq_put_decimal_ll(m
, "\nNSpid:\t", nr
);
1772 /* If nr is non-zero it means that 'pid' is valid and that
1773 * ns, i.e. the pid namespace associated with the procfs
1774 * instance, is in the pid namespace hierarchy of pid.
1775 * Start at one below the already printed level.
1777 for (i
= ns
->level
+ 1; i
<= pid
->level
; i
++)
1778 seq_put_decimal_ll(m
, "\t", pid
->numbers
[i
].nr
);
1786 * Poll support for process exit notification.
1788 static __poll_t
pidfd_poll(struct file
*file
, struct poll_table_struct
*pts
)
1790 struct task_struct
*task
;
1791 struct pid
*pid
= file
->private_data
;
1792 __poll_t poll_flags
= 0;
1794 poll_wait(file
, &pid
->wait_pidfd
, pts
);
1797 task
= pid_task(pid
, PIDTYPE_PID
);
1799 * Inform pollers only when the whole thread group exits.
1800 * If the thread group leader exits before all other threads in the
1801 * group, then poll(2) should block, similar to the wait(2) family.
1803 if (!task
|| (task
->exit_state
&& thread_group_empty(task
)))
1804 poll_flags
= EPOLLIN
| EPOLLRDNORM
;
1810 const struct file_operations pidfd_fops
= {
1811 .release
= pidfd_release
,
1813 #ifdef CONFIG_PROC_FS
1814 .show_fdinfo
= pidfd_show_fdinfo
,
1818 static void __delayed_free_task(struct rcu_head
*rhp
)
1820 struct task_struct
*tsk
= container_of(rhp
, struct task_struct
, rcu
);
1825 static __always_inline
void delayed_free_task(struct task_struct
*tsk
)
1827 if (IS_ENABLED(CONFIG_MEMCG
))
1828 call_rcu(&tsk
->rcu
, __delayed_free_task
);
1834 * This creates a new process as a copy of the old one,
1835 * but does not actually start it yet.
1837 * It copies the registers, and all the appropriate
1838 * parts of the process environment (as per the clone
1839 * flags). The actual kick-off is left to the caller.
1841 static __latent_entropy
struct task_struct
*copy_process(
1845 struct kernel_clone_args
*args
)
1847 int pidfd
= -1, retval
;
1848 struct task_struct
*p
;
1849 struct multiprocess_signals delayed
;
1850 struct file
*pidfile
= NULL
;
1851 u64 clone_flags
= args
->flags
;
1852 struct nsproxy
*nsp
= current
->nsproxy
;
1855 * Don't allow sharing the root directory with processes in a different
1858 if ((clone_flags
& (CLONE_NEWNS
|CLONE_FS
)) == (CLONE_NEWNS
|CLONE_FS
))
1859 return ERR_PTR(-EINVAL
);
1861 if ((clone_flags
& (CLONE_NEWUSER
|CLONE_FS
)) == (CLONE_NEWUSER
|CLONE_FS
))
1862 return ERR_PTR(-EINVAL
);
1865 * Thread groups must share signals as well, and detached threads
1866 * can only be started up within the thread group.
1868 if ((clone_flags
& CLONE_THREAD
) && !(clone_flags
& CLONE_SIGHAND
))
1869 return ERR_PTR(-EINVAL
);
1872 * Shared signal handlers imply shared VM. By way of the above,
1873 * thread groups also imply shared VM. Blocking this case allows
1874 * for various simplifications in other code.
1876 if ((clone_flags
& CLONE_SIGHAND
) && !(clone_flags
& CLONE_VM
))
1877 return ERR_PTR(-EINVAL
);
1880 * Siblings of global init remain as zombies on exit since they are
1881 * not reaped by their parent (swapper). To solve this and to avoid
1882 * multi-rooted process trees, prevent global and container-inits
1883 * from creating siblings.
1885 if ((clone_flags
& CLONE_PARENT
) &&
1886 current
->signal
->flags
& SIGNAL_UNKILLABLE
)
1887 return ERR_PTR(-EINVAL
);
1890 * If the new process will be in a different pid or user namespace
1891 * do not allow it to share a thread group with the forking task.
1893 if (clone_flags
& CLONE_THREAD
) {
1894 if ((clone_flags
& (CLONE_NEWUSER
| CLONE_NEWPID
)) ||
1895 (task_active_pid_ns(current
) != nsp
->pid_ns_for_children
))
1896 return ERR_PTR(-EINVAL
);
1900 * If the new process will be in a different time namespace
1901 * do not allow it to share VM or a thread group with the forking task.
1903 if (clone_flags
& (CLONE_THREAD
| CLONE_VM
)) {
1904 if (nsp
->time_ns
!= nsp
->time_ns_for_children
)
1905 return ERR_PTR(-EINVAL
);
1908 if (clone_flags
& CLONE_PIDFD
) {
1910 * - CLONE_DETACHED is blocked so that we can potentially
1911 * reuse it later for CLONE_PIDFD.
1912 * - CLONE_THREAD is blocked until someone really needs it.
1914 if (clone_flags
& (CLONE_DETACHED
| CLONE_THREAD
))
1915 return ERR_PTR(-EINVAL
);
1919 * Force any signals received before this point to be delivered
1920 * before the fork happens. Collect up signals sent to multiple
1921 * processes that happen during the fork and delay them so that
1922 * they appear to happen after the fork.
1924 sigemptyset(&delayed
.signal
);
1925 INIT_HLIST_NODE(&delayed
.node
);
1927 spin_lock_irq(¤t
->sighand
->siglock
);
1928 if (!(clone_flags
& CLONE_THREAD
))
1929 hlist_add_head(&delayed
.node
, ¤t
->signal
->multiprocess
);
1930 recalc_sigpending();
1931 spin_unlock_irq(¤t
->sighand
->siglock
);
1932 retval
= -ERESTARTNOINTR
;
1933 if (signal_pending(current
))
1937 p
= dup_task_struct(current
, node
);
1942 * This _must_ happen before we call free_task(), i.e. before we jump
1943 * to any of the bad_fork_* labels. This is to avoid freeing
1944 * p->set_child_tid which is (ab)used as a kthread's data pointer for
1945 * kernel threads (PF_KTHREAD).
1947 p
->set_child_tid
= (clone_flags
& CLONE_CHILD_SETTID
) ? args
->child_tid
: NULL
;
1949 * Clear TID on mm_release()?
1951 p
->clear_child_tid
= (clone_flags
& CLONE_CHILD_CLEARTID
) ? args
->child_tid
: NULL
;
1953 ftrace_graph_init_task(p
);
1955 rt_mutex_init_task(p
);
1957 lockdep_assert_irqs_enabled();
1958 #ifdef CONFIG_PROVE_LOCKING
1959 DEBUG_LOCKS_WARN_ON(!p
->softirqs_enabled
);
1962 if (atomic_read(&p
->real_cred
->user
->processes
) >=
1963 task_rlimit(p
, RLIMIT_NPROC
)) {
1964 if (p
->real_cred
->user
!= INIT_USER
&&
1965 !capable(CAP_SYS_RESOURCE
) && !capable(CAP_SYS_ADMIN
))
1968 current
->flags
&= ~PF_NPROC_EXCEEDED
;
1970 retval
= copy_creds(p
, clone_flags
);
1975 * If multiple threads are within copy_process(), then this check
1976 * triggers too late. This doesn't hurt, the check is only there
1977 * to stop root fork bombs.
1980 if (data_race(nr_threads
>= max_threads
))
1981 goto bad_fork_cleanup_count
;
1983 delayacct_tsk_init(p
); /* Must remain after dup_task_struct() */
1984 p
->flags
&= ~(PF_SUPERPRIV
| PF_WQ_WORKER
| PF_IDLE
);
1985 p
->flags
|= PF_FORKNOEXEC
;
1986 INIT_LIST_HEAD(&p
->children
);
1987 INIT_LIST_HEAD(&p
->sibling
);
1988 rcu_copy_process(p
);
1989 p
->vfork_done
= NULL
;
1990 spin_lock_init(&p
->alloc_lock
);
1992 init_sigpending(&p
->pending
);
1994 p
->utime
= p
->stime
= p
->gtime
= 0;
1995 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
1996 p
->utimescaled
= p
->stimescaled
= 0;
1998 prev_cputime_init(&p
->prev_cputime
);
2000 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
2001 seqcount_init(&p
->vtime
.seqcount
);
2002 p
->vtime
.starttime
= 0;
2003 p
->vtime
.state
= VTIME_INACTIVE
;
2006 #if defined(SPLIT_RSS_COUNTING)
2007 memset(&p
->rss_stat
, 0, sizeof(p
->rss_stat
));
2010 p
->default_timer_slack_ns
= current
->timer_slack_ns
;
2016 task_io_accounting_init(&p
->ioac
);
2017 acct_clear_integrals(p
);
2019 posix_cputimers_init(&p
->posix_cputimers
);
2021 p
->io_context
= NULL
;
2022 audit_set_context(p
, NULL
);
2025 p
->mempolicy
= mpol_dup(p
->mempolicy
);
2026 if (IS_ERR(p
->mempolicy
)) {
2027 retval
= PTR_ERR(p
->mempolicy
);
2028 p
->mempolicy
= NULL
;
2029 goto bad_fork_cleanup_threadgroup_lock
;
2032 #ifdef CONFIG_CPUSETS
2033 p
->cpuset_mem_spread_rotor
= NUMA_NO_NODE
;
2034 p
->cpuset_slab_spread_rotor
= NUMA_NO_NODE
;
2035 seqcount_init(&p
->mems_allowed_seq
);
2037 #ifdef CONFIG_TRACE_IRQFLAGS
2039 p
->hardirq_enable_ip
= 0;
2040 p
->hardirq_enable_event
= 0;
2041 p
->hardirq_disable_ip
= _THIS_IP_
;
2042 p
->hardirq_disable_event
= 0;
2043 p
->softirqs_enabled
= 1;
2044 p
->softirq_enable_ip
= _THIS_IP_
;
2045 p
->softirq_enable_event
= 0;
2046 p
->softirq_disable_ip
= 0;
2047 p
->softirq_disable_event
= 0;
2048 p
->softirq_context
= 0;
2051 p
->pagefault_disabled
= 0;
2053 #ifdef CONFIG_LOCKDEP
2054 lockdep_init_task(p
);
2057 #ifdef CONFIG_DEBUG_MUTEXES
2058 p
->blocked_on
= NULL
; /* not blocked yet */
2060 #ifdef CONFIG_BCACHE
2061 p
->sequential_io
= 0;
2062 p
->sequential_io_avg
= 0;
2065 /* Perform scheduler related setup. Assign this task to a CPU. */
2066 retval
= sched_fork(clone_flags
, p
);
2068 goto bad_fork_cleanup_policy
;
2070 retval
= perf_event_init_task(p
);
2072 goto bad_fork_cleanup_policy
;
2073 retval
= audit_alloc(p
);
2075 goto bad_fork_cleanup_perf
;
2076 /* copy all the process information */
2078 retval
= security_task_alloc(p
, clone_flags
);
2080 goto bad_fork_cleanup_audit
;
2081 retval
= copy_semundo(clone_flags
, p
);
2083 goto bad_fork_cleanup_security
;
2084 retval
= copy_files(clone_flags
, p
);
2086 goto bad_fork_cleanup_semundo
;
2087 retval
= copy_fs(clone_flags
, p
);
2089 goto bad_fork_cleanup_files
;
2090 retval
= copy_sighand(clone_flags
, p
);
2092 goto bad_fork_cleanup_fs
;
2093 retval
= copy_signal(clone_flags
, p
);
2095 goto bad_fork_cleanup_sighand
;
2096 retval
= copy_mm(clone_flags
, p
);
2098 goto bad_fork_cleanup_signal
;
2099 retval
= copy_namespaces(clone_flags
, p
);
2101 goto bad_fork_cleanup_mm
;
2102 retval
= copy_io(clone_flags
, p
);
2104 goto bad_fork_cleanup_namespaces
;
2105 retval
= copy_thread_tls(clone_flags
, args
->stack
, args
->stack_size
, p
,
2108 goto bad_fork_cleanup_io
;
2110 stackleak_task_init(p
);
2112 if (pid
!= &init_struct_pid
) {
2113 pid
= alloc_pid(p
->nsproxy
->pid_ns_for_children
, args
->set_tid
,
2114 args
->set_tid_size
);
2116 retval
= PTR_ERR(pid
);
2117 goto bad_fork_cleanup_thread
;
2122 * This has to happen after we've potentially unshared the file
2123 * descriptor table (so that the pidfd doesn't leak into the child
2124 * if the fd table isn't shared).
2126 if (clone_flags
& CLONE_PIDFD
) {
2127 retval
= get_unused_fd_flags(O_RDWR
| O_CLOEXEC
);
2129 goto bad_fork_free_pid
;
2133 pidfile
= anon_inode_getfile("[pidfd]", &pidfd_fops
, pid
,
2134 O_RDWR
| O_CLOEXEC
);
2135 if (IS_ERR(pidfile
)) {
2136 put_unused_fd(pidfd
);
2137 retval
= PTR_ERR(pidfile
);
2138 goto bad_fork_free_pid
;
2140 get_pid(pid
); /* held by pidfile now */
2142 retval
= put_user(pidfd
, args
->pidfd
);
2144 goto bad_fork_put_pidfd
;
2153 * sigaltstack should be cleared when sharing the same VM
2155 if ((clone_flags
& (CLONE_VM
|CLONE_VFORK
)) == CLONE_VM
)
2159 * Syscall tracing and stepping should be turned off in the
2160 * child regardless of CLONE_PTRACE.
2162 user_disable_single_step(p
);
2163 clear_tsk_thread_flag(p
, TIF_SYSCALL_TRACE
);
2164 #ifdef TIF_SYSCALL_EMU
2165 clear_tsk_thread_flag(p
, TIF_SYSCALL_EMU
);
2167 clear_tsk_latency_tracing(p
);
2169 /* ok, now we should be set up.. */
2170 p
->pid
= pid_nr(pid
);
2171 if (clone_flags
& CLONE_THREAD
) {
2172 p
->exit_signal
= -1;
2173 p
->group_leader
= current
->group_leader
;
2174 p
->tgid
= current
->tgid
;
2176 if (clone_flags
& CLONE_PARENT
)
2177 p
->exit_signal
= current
->group_leader
->exit_signal
;
2179 p
->exit_signal
= args
->exit_signal
;
2180 p
->group_leader
= p
;
2185 p
->nr_dirtied_pause
= 128 >> (PAGE_SHIFT
- 10);
2186 p
->dirty_paused_when
= 0;
2188 p
->pdeath_signal
= 0;
2189 INIT_LIST_HEAD(&p
->thread_group
);
2190 p
->task_works
= NULL
;
2193 * Ensure that the cgroup subsystem policies allow the new process to be
2194 * forked. It should be noted the the new process's css_set can be changed
2195 * between here and cgroup_post_fork() if an organisation operation is in
2198 retval
= cgroup_can_fork(p
, args
);
2200 goto bad_fork_put_pidfd
;
2203 * From this point on we must avoid any synchronous user-space
2204 * communication until we take the tasklist-lock. In particular, we do
2205 * not want user-space to be able to predict the process start-time by
2206 * stalling fork(2) after we recorded the start_time but before it is
2207 * visible to the system.
2210 p
->start_time
= ktime_get_ns();
2211 p
->start_boottime
= ktime_get_boottime_ns();
2214 * Make it visible to the rest of the system, but dont wake it up yet.
2215 * Need tasklist lock for parent etc handling!
2217 write_lock_irq(&tasklist_lock
);
2219 /* CLONE_PARENT re-uses the old parent */
2220 if (clone_flags
& (CLONE_PARENT
|CLONE_THREAD
)) {
2221 p
->real_parent
= current
->real_parent
;
2222 p
->parent_exec_id
= current
->parent_exec_id
;
2224 p
->real_parent
= current
;
2225 p
->parent_exec_id
= current
->self_exec_id
;
2228 klp_copy_process(p
);
2230 spin_lock(¤t
->sighand
->siglock
);
2233 * Copy seccomp details explicitly here, in case they were changed
2234 * before holding sighand lock.
2238 rseq_fork(p
, clone_flags
);
2240 /* Don't start children in a dying pid namespace */
2241 if (unlikely(!(ns_of_pid(pid
)->pid_allocated
& PIDNS_ADDING
))) {
2243 goto bad_fork_cancel_cgroup
;
2246 /* Let kill terminate clone/fork in the middle */
2247 if (fatal_signal_pending(current
)) {
2249 goto bad_fork_cancel_cgroup
;
2252 /* past the last point of failure */
2254 fd_install(pidfd
, pidfile
);
2256 init_task_pid_links(p
);
2257 if (likely(p
->pid
)) {
2258 ptrace_init_task(p
, (clone_flags
& CLONE_PTRACE
) || trace
);
2260 init_task_pid(p
, PIDTYPE_PID
, pid
);
2261 if (thread_group_leader(p
)) {
2262 init_task_pid(p
, PIDTYPE_TGID
, pid
);
2263 init_task_pid(p
, PIDTYPE_PGID
, task_pgrp(current
));
2264 init_task_pid(p
, PIDTYPE_SID
, task_session(current
));
2266 if (is_child_reaper(pid
)) {
2267 ns_of_pid(pid
)->child_reaper
= p
;
2268 p
->signal
->flags
|= SIGNAL_UNKILLABLE
;
2270 p
->signal
->shared_pending
.signal
= delayed
.signal
;
2271 p
->signal
->tty
= tty_kref_get(current
->signal
->tty
);
2273 * Inherit has_child_subreaper flag under the same
2274 * tasklist_lock with adding child to the process tree
2275 * for propagate_has_child_subreaper optimization.
2277 p
->signal
->has_child_subreaper
= p
->real_parent
->signal
->has_child_subreaper
||
2278 p
->real_parent
->signal
->is_child_subreaper
;
2279 list_add_tail(&p
->sibling
, &p
->real_parent
->children
);
2280 list_add_tail_rcu(&p
->tasks
, &init_task
.tasks
);
2281 attach_pid(p
, PIDTYPE_TGID
);
2282 attach_pid(p
, PIDTYPE_PGID
);
2283 attach_pid(p
, PIDTYPE_SID
);
2284 __this_cpu_inc(process_counts
);
2286 current
->signal
->nr_threads
++;
2287 atomic_inc(¤t
->signal
->live
);
2288 refcount_inc(¤t
->signal
->sigcnt
);
2289 task_join_group_stop(p
);
2290 list_add_tail_rcu(&p
->thread_group
,
2291 &p
->group_leader
->thread_group
);
2292 list_add_tail_rcu(&p
->thread_node
,
2293 &p
->signal
->thread_head
);
2295 attach_pid(p
, PIDTYPE_PID
);
2299 hlist_del_init(&delayed
.node
);
2300 spin_unlock(¤t
->sighand
->siglock
);
2301 syscall_tracepoint_update(p
);
2302 write_unlock_irq(&tasklist_lock
);
2304 proc_fork_connector(p
);
2305 cgroup_post_fork(p
, args
);
2308 trace_task_newtask(p
, clone_flags
);
2309 uprobe_copy_process(p
, clone_flags
);
2313 bad_fork_cancel_cgroup
:
2314 spin_unlock(¤t
->sighand
->siglock
);
2315 write_unlock_irq(&tasklist_lock
);
2316 cgroup_cancel_fork(p
, args
);
2318 if (clone_flags
& CLONE_PIDFD
) {
2320 put_unused_fd(pidfd
);
2323 if (pid
!= &init_struct_pid
)
2325 bad_fork_cleanup_thread
:
2327 bad_fork_cleanup_io
:
2330 bad_fork_cleanup_namespaces
:
2331 exit_task_namespaces(p
);
2332 bad_fork_cleanup_mm
:
2334 mm_clear_owner(p
->mm
, p
);
2337 bad_fork_cleanup_signal
:
2338 if (!(clone_flags
& CLONE_THREAD
))
2339 free_signal_struct(p
->signal
);
2340 bad_fork_cleanup_sighand
:
2341 __cleanup_sighand(p
->sighand
);
2342 bad_fork_cleanup_fs
:
2343 exit_fs(p
); /* blocking */
2344 bad_fork_cleanup_files
:
2345 exit_files(p
); /* blocking */
2346 bad_fork_cleanup_semundo
:
2348 bad_fork_cleanup_security
:
2349 security_task_free(p
);
2350 bad_fork_cleanup_audit
:
2352 bad_fork_cleanup_perf
:
2353 perf_event_free_task(p
);
2354 bad_fork_cleanup_policy
:
2355 lockdep_free_task(p
);
2357 mpol_put(p
->mempolicy
);
2358 bad_fork_cleanup_threadgroup_lock
:
2360 delayacct_tsk_free(p
);
2361 bad_fork_cleanup_count
:
2362 atomic_dec(&p
->cred
->user
->processes
);
2365 p
->state
= TASK_DEAD
;
2367 delayed_free_task(p
);
2369 spin_lock_irq(¤t
->sighand
->siglock
);
2370 hlist_del_init(&delayed
.node
);
2371 spin_unlock_irq(¤t
->sighand
->siglock
);
2372 return ERR_PTR(retval
);
2375 static inline void init_idle_pids(struct task_struct
*idle
)
2379 for (type
= PIDTYPE_PID
; type
< PIDTYPE_MAX
; ++type
) {
2380 INIT_HLIST_NODE(&idle
->pid_links
[type
]); /* not really needed */
2381 init_task_pid(idle
, type
, &init_struct_pid
);
2385 struct task_struct
*fork_idle(int cpu
)
2387 struct task_struct
*task
;
2388 struct kernel_clone_args args
= {
2392 task
= copy_process(&init_struct_pid
, 0, cpu_to_node(cpu
), &args
);
2393 if (!IS_ERR(task
)) {
2394 init_idle_pids(task
);
2395 init_idle(task
, cpu
);
2401 struct mm_struct
*copy_init_mm(void)
2403 return dup_mm(NULL
, &init_mm
);
2407 * Ok, this is the main fork-routine.
2409 * It copies the process, and if successful kick-starts
2410 * it and waits for it to finish using the VM if required.
2412 * args->exit_signal is expected to be checked for sanity by the caller.
2414 long _do_fork(struct kernel_clone_args
*args
)
2416 u64 clone_flags
= args
->flags
;
2417 struct completion vfork
;
2419 struct task_struct
*p
;
2424 * Determine whether and which event to report to ptracer. When
2425 * called from kernel_thread or CLONE_UNTRACED is explicitly
2426 * requested, no event is reported; otherwise, report if the event
2427 * for the type of forking is enabled.
2429 if (!(clone_flags
& CLONE_UNTRACED
)) {
2430 if (clone_flags
& CLONE_VFORK
)
2431 trace
= PTRACE_EVENT_VFORK
;
2432 else if (args
->exit_signal
!= SIGCHLD
)
2433 trace
= PTRACE_EVENT_CLONE
;
2435 trace
= PTRACE_EVENT_FORK
;
2437 if (likely(!ptrace_event_enabled(current
, trace
)))
2441 p
= copy_process(NULL
, trace
, NUMA_NO_NODE
, args
);
2442 add_latent_entropy();
2448 * Do this prior waking up the new thread - the thread pointer
2449 * might get invalid after that point, if the thread exits quickly.
2451 trace_sched_process_fork(current
, p
);
2453 pid
= get_task_pid(p
, PIDTYPE_PID
);
2456 if (clone_flags
& CLONE_PARENT_SETTID
)
2457 put_user(nr
, args
->parent_tid
);
2459 if (clone_flags
& CLONE_VFORK
) {
2460 p
->vfork_done
= &vfork
;
2461 init_completion(&vfork
);
2465 wake_up_new_task(p
);
2467 /* forking complete and child started to run, tell ptracer */
2468 if (unlikely(trace
))
2469 ptrace_event_pid(trace
, pid
);
2471 if (clone_flags
& CLONE_VFORK
) {
2472 if (!wait_for_vfork_done(p
, &vfork
))
2473 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE
, pid
);
2480 bool legacy_clone_args_valid(const struct kernel_clone_args
*kargs
)
2482 /* clone(CLONE_PIDFD) uses parent_tidptr to return a pidfd */
2483 if ((kargs
->flags
& CLONE_PIDFD
) &&
2484 (kargs
->flags
& CLONE_PARENT_SETTID
))
2490 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
2491 /* For compatibility with architectures that call do_fork directly rather than
2492 * using the syscall entry points below. */
2493 long do_fork(unsigned long clone_flags
,
2494 unsigned long stack_start
,
2495 unsigned long stack_size
,
2496 int __user
*parent_tidptr
,
2497 int __user
*child_tidptr
)
2499 struct kernel_clone_args args
= {
2500 .flags
= (lower_32_bits(clone_flags
) & ~CSIGNAL
),
2501 .pidfd
= parent_tidptr
,
2502 .child_tid
= child_tidptr
,
2503 .parent_tid
= parent_tidptr
,
2504 .exit_signal
= (lower_32_bits(clone_flags
) & CSIGNAL
),
2505 .stack
= stack_start
,
2506 .stack_size
= stack_size
,
2509 if (!legacy_clone_args_valid(&args
))
2512 return _do_fork(&args
);
2517 * Create a kernel thread.
2519 pid_t
kernel_thread(int (*fn
)(void *), void *arg
, unsigned long flags
)
2521 struct kernel_clone_args args
= {
2522 .flags
= ((lower_32_bits(flags
) | CLONE_VM
|
2523 CLONE_UNTRACED
) & ~CSIGNAL
),
2524 .exit_signal
= (lower_32_bits(flags
) & CSIGNAL
),
2525 .stack
= (unsigned long)fn
,
2526 .stack_size
= (unsigned long)arg
,
2529 return _do_fork(&args
);
2532 #ifdef __ARCH_WANT_SYS_FORK
2533 SYSCALL_DEFINE0(fork
)
2536 struct kernel_clone_args args
= {
2537 .exit_signal
= SIGCHLD
,
2540 return _do_fork(&args
);
2542 /* can not support in nommu mode */
2548 #ifdef __ARCH_WANT_SYS_VFORK
2549 SYSCALL_DEFINE0(vfork
)
2551 struct kernel_clone_args args
= {
2552 .flags
= CLONE_VFORK
| CLONE_VM
,
2553 .exit_signal
= SIGCHLD
,
2556 return _do_fork(&args
);
2560 #ifdef __ARCH_WANT_SYS_CLONE
2561 #ifdef CONFIG_CLONE_BACKWARDS
2562 SYSCALL_DEFINE5(clone
, unsigned long, clone_flags
, unsigned long, newsp
,
2563 int __user
*, parent_tidptr
,
2565 int __user
*, child_tidptr
)
2566 #elif defined(CONFIG_CLONE_BACKWARDS2)
2567 SYSCALL_DEFINE5(clone
, unsigned long, newsp
, unsigned long, clone_flags
,
2568 int __user
*, parent_tidptr
,
2569 int __user
*, child_tidptr
,
2571 #elif defined(CONFIG_CLONE_BACKWARDS3)
2572 SYSCALL_DEFINE6(clone
, unsigned long, clone_flags
, unsigned long, newsp
,
2574 int __user
*, parent_tidptr
,
2575 int __user
*, child_tidptr
,
2578 SYSCALL_DEFINE5(clone
, unsigned long, clone_flags
, unsigned long, newsp
,
2579 int __user
*, parent_tidptr
,
2580 int __user
*, child_tidptr
,
2584 struct kernel_clone_args args
= {
2585 .flags
= (lower_32_bits(clone_flags
) & ~CSIGNAL
),
2586 .pidfd
= parent_tidptr
,
2587 .child_tid
= child_tidptr
,
2588 .parent_tid
= parent_tidptr
,
2589 .exit_signal
= (lower_32_bits(clone_flags
) & CSIGNAL
),
2594 if (!legacy_clone_args_valid(&args
))
2597 return _do_fork(&args
);
2601 #ifdef __ARCH_WANT_SYS_CLONE3
2604 * copy_thread implementations handle CLONE_SETTLS by reading the TLS value from
2605 * the registers containing the syscall arguments for clone. This doesn't work
2606 * with clone3 since the TLS value is passed in clone_args instead.
2608 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
2609 #error clone3 requires copy_thread_tls support in arch
2612 noinline
static int copy_clone_args_from_user(struct kernel_clone_args
*kargs
,
2613 struct clone_args __user
*uargs
,
2617 struct clone_args args
;
2618 pid_t
*kset_tid
= kargs
->set_tid
;
2620 BUILD_BUG_ON(offsetofend(struct clone_args
, tls
) !=
2621 CLONE_ARGS_SIZE_VER0
);
2622 BUILD_BUG_ON(offsetofend(struct clone_args
, set_tid_size
) !=
2623 CLONE_ARGS_SIZE_VER1
);
2624 BUILD_BUG_ON(offsetofend(struct clone_args
, cgroup
) !=
2625 CLONE_ARGS_SIZE_VER2
);
2626 BUILD_BUG_ON(sizeof(struct clone_args
) != CLONE_ARGS_SIZE_VER2
);
2628 if (unlikely(usize
> PAGE_SIZE
))
2630 if (unlikely(usize
< CLONE_ARGS_SIZE_VER0
))
2633 err
= copy_struct_from_user(&args
, sizeof(args
), uargs
, usize
);
2637 if (unlikely(args
.set_tid_size
> MAX_PID_NS_LEVEL
))
2640 if (unlikely(!args
.set_tid
&& args
.set_tid_size
> 0))
2643 if (unlikely(args
.set_tid
&& args
.set_tid_size
== 0))
2647 * Verify that higher 32bits of exit_signal are unset and that
2648 * it is a valid signal
2650 if (unlikely((args
.exit_signal
& ~((u64
)CSIGNAL
)) ||
2651 !valid_signal(args
.exit_signal
)))
2654 if ((args
.flags
& CLONE_INTO_CGROUP
) &&
2655 (args
.cgroup
> INT_MAX
|| usize
< CLONE_ARGS_SIZE_VER2
))
2658 *kargs
= (struct kernel_clone_args
){
2659 .flags
= args
.flags
,
2660 .pidfd
= u64_to_user_ptr(args
.pidfd
),
2661 .child_tid
= u64_to_user_ptr(args
.child_tid
),
2662 .parent_tid
= u64_to_user_ptr(args
.parent_tid
),
2663 .exit_signal
= args
.exit_signal
,
2664 .stack
= args
.stack
,
2665 .stack_size
= args
.stack_size
,
2667 .set_tid_size
= args
.set_tid_size
,
2668 .cgroup
= args
.cgroup
,
2672 copy_from_user(kset_tid
, u64_to_user_ptr(args
.set_tid
),
2673 (kargs
->set_tid_size
* sizeof(pid_t
))))
2676 kargs
->set_tid
= kset_tid
;
2682 * clone3_stack_valid - check and prepare stack
2683 * @kargs: kernel clone args
2685 * Verify that the stack arguments userspace gave us are sane.
2686 * In addition, set the stack direction for userspace since it's easy for us to
2689 static inline bool clone3_stack_valid(struct kernel_clone_args
*kargs
)
2691 if (kargs
->stack
== 0) {
2692 if (kargs
->stack_size
> 0)
2695 if (kargs
->stack_size
== 0)
2698 if (!access_ok((void __user
*)kargs
->stack
, kargs
->stack_size
))
2701 #if !defined(CONFIG_STACK_GROWSUP) && !defined(CONFIG_IA64)
2702 kargs
->stack
+= kargs
->stack_size
;
2709 static bool clone3_args_valid(struct kernel_clone_args
*kargs
)
2711 /* Verify that no unknown flags are passed along. */
2713 ~(CLONE_LEGACY_FLAGS
| CLONE_CLEAR_SIGHAND
| CLONE_INTO_CGROUP
))
2717 * - make the CLONE_DETACHED bit reuseable for clone3
2718 * - make the CSIGNAL bits reuseable for clone3
2720 if (kargs
->flags
& (CLONE_DETACHED
| CSIGNAL
))
2723 if ((kargs
->flags
& (CLONE_SIGHAND
| CLONE_CLEAR_SIGHAND
)) ==
2724 (CLONE_SIGHAND
| CLONE_CLEAR_SIGHAND
))
2727 if ((kargs
->flags
& (CLONE_THREAD
| CLONE_PARENT
)) &&
2731 if (!clone3_stack_valid(kargs
))
2738 * clone3 - create a new process with specific properties
2739 * @uargs: argument structure
2740 * @size: size of @uargs
2742 * clone3() is the extensible successor to clone()/clone2().
2743 * It takes a struct as argument that is versioned by its size.
2745 * Return: On success, a positive PID for the child process.
2746 * On error, a negative errno number.
2748 SYSCALL_DEFINE2(clone3
, struct clone_args __user
*, uargs
, size_t, size
)
2752 struct kernel_clone_args kargs
;
2753 pid_t set_tid
[MAX_PID_NS_LEVEL
];
2755 kargs
.set_tid
= set_tid
;
2757 err
= copy_clone_args_from_user(&kargs
, uargs
, size
);
2761 if (!clone3_args_valid(&kargs
))
2764 return _do_fork(&kargs
);
2768 void walk_process_tree(struct task_struct
*top
, proc_visitor visitor
, void *data
)
2770 struct task_struct
*leader
, *parent
, *child
;
2773 read_lock(&tasklist_lock
);
2774 leader
= top
= top
->group_leader
;
2776 for_each_thread(leader
, parent
) {
2777 list_for_each_entry(child
, &parent
->children
, sibling
) {
2778 res
= visitor(child
, data
);
2790 if (leader
!= top
) {
2792 parent
= child
->real_parent
;
2793 leader
= parent
->group_leader
;
2797 read_unlock(&tasklist_lock
);
2800 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
2801 #define ARCH_MIN_MMSTRUCT_ALIGN 0
2804 static void sighand_ctor(void *data
)
2806 struct sighand_struct
*sighand
= data
;
2808 spin_lock_init(&sighand
->siglock
);
2809 init_waitqueue_head(&sighand
->signalfd_wqh
);
2812 void __init
proc_caches_init(void)
2814 unsigned int mm_size
;
2816 sighand_cachep
= kmem_cache_create("sighand_cache",
2817 sizeof(struct sighand_struct
), 0,
2818 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
|SLAB_TYPESAFE_BY_RCU
|
2819 SLAB_ACCOUNT
, sighand_ctor
);
2820 signal_cachep
= kmem_cache_create("signal_cache",
2821 sizeof(struct signal_struct
), 0,
2822 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
|SLAB_ACCOUNT
,
2824 files_cachep
= kmem_cache_create("files_cache",
2825 sizeof(struct files_struct
), 0,
2826 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
|SLAB_ACCOUNT
,
2828 fs_cachep
= kmem_cache_create("fs_cache",
2829 sizeof(struct fs_struct
), 0,
2830 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
|SLAB_ACCOUNT
,
2834 * The mm_cpumask is located at the end of mm_struct, and is
2835 * dynamically sized based on the maximum CPU number this system
2836 * can have, taking hotplug into account (nr_cpu_ids).
2838 mm_size
= sizeof(struct mm_struct
) + cpumask_size();
2840 mm_cachep
= kmem_cache_create_usercopy("mm_struct",
2841 mm_size
, ARCH_MIN_MMSTRUCT_ALIGN
,
2842 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
|SLAB_ACCOUNT
,
2843 offsetof(struct mm_struct
, saved_auxv
),
2844 sizeof_field(struct mm_struct
, saved_auxv
),
2846 vm_area_cachep
= KMEM_CACHE(vm_area_struct
, SLAB_PANIC
|SLAB_ACCOUNT
);
2848 nsproxy_cache_init();
2852 * Check constraints on flags passed to the unshare system call.
2854 static int check_unshare_flags(unsigned long unshare_flags
)
2856 if (unshare_flags
& ~(CLONE_THREAD
|CLONE_FS
|CLONE_NEWNS
|CLONE_SIGHAND
|
2857 CLONE_VM
|CLONE_FILES
|CLONE_SYSVSEM
|
2858 CLONE_NEWUTS
|CLONE_NEWIPC
|CLONE_NEWNET
|
2859 CLONE_NEWUSER
|CLONE_NEWPID
|CLONE_NEWCGROUP
|
2863 * Not implemented, but pretend it works if there is nothing
2864 * to unshare. Note that unsharing the address space or the
2865 * signal handlers also need to unshare the signal queues (aka
2868 if (unshare_flags
& (CLONE_THREAD
| CLONE_SIGHAND
| CLONE_VM
)) {
2869 if (!thread_group_empty(current
))
2872 if (unshare_flags
& (CLONE_SIGHAND
| CLONE_VM
)) {
2873 if (refcount_read(¤t
->sighand
->count
) > 1)
2876 if (unshare_flags
& CLONE_VM
) {
2877 if (!current_is_single_threaded())
2885 * Unshare the filesystem structure if it is being shared
2887 static int unshare_fs(unsigned long unshare_flags
, struct fs_struct
**new_fsp
)
2889 struct fs_struct
*fs
= current
->fs
;
2891 if (!(unshare_flags
& CLONE_FS
) || !fs
)
2894 /* don't need lock here; in the worst case we'll do useless copy */
2898 *new_fsp
= copy_fs_struct(fs
);
2906 * Unshare file descriptor table if it is being shared
2908 static int unshare_fd(unsigned long unshare_flags
, struct files_struct
**new_fdp
)
2910 struct files_struct
*fd
= current
->files
;
2913 if ((unshare_flags
& CLONE_FILES
) &&
2914 (fd
&& atomic_read(&fd
->count
) > 1)) {
2915 *new_fdp
= dup_fd(fd
, &error
);
2924 * unshare allows a process to 'unshare' part of the process
2925 * context which was originally shared using clone. copy_*
2926 * functions used by do_fork() cannot be used here directly
2927 * because they modify an inactive task_struct that is being
2928 * constructed. Here we are modifying the current, active,
2931 int ksys_unshare(unsigned long unshare_flags
)
2933 struct fs_struct
*fs
, *new_fs
= NULL
;
2934 struct files_struct
*fd
, *new_fd
= NULL
;
2935 struct cred
*new_cred
= NULL
;
2936 struct nsproxy
*new_nsproxy
= NULL
;
2941 * If unsharing a user namespace must also unshare the thread group
2942 * and unshare the filesystem root and working directories.
2944 if (unshare_flags
& CLONE_NEWUSER
)
2945 unshare_flags
|= CLONE_THREAD
| CLONE_FS
;
2947 * If unsharing vm, must also unshare signal handlers.
2949 if (unshare_flags
& CLONE_VM
)
2950 unshare_flags
|= CLONE_SIGHAND
;
2952 * If unsharing a signal handlers, must also unshare the signal queues.
2954 if (unshare_flags
& CLONE_SIGHAND
)
2955 unshare_flags
|= CLONE_THREAD
;
2957 * If unsharing namespace, must also unshare filesystem information.
2959 if (unshare_flags
& CLONE_NEWNS
)
2960 unshare_flags
|= CLONE_FS
;
2962 err
= check_unshare_flags(unshare_flags
);
2964 goto bad_unshare_out
;
2966 * CLONE_NEWIPC must also detach from the undolist: after switching
2967 * to a new ipc namespace, the semaphore arrays from the old
2968 * namespace are unreachable.
2970 if (unshare_flags
& (CLONE_NEWIPC
|CLONE_SYSVSEM
))
2972 err
= unshare_fs(unshare_flags
, &new_fs
);
2974 goto bad_unshare_out
;
2975 err
= unshare_fd(unshare_flags
, &new_fd
);
2977 goto bad_unshare_cleanup_fs
;
2978 err
= unshare_userns(unshare_flags
, &new_cred
);
2980 goto bad_unshare_cleanup_fd
;
2981 err
= unshare_nsproxy_namespaces(unshare_flags
, &new_nsproxy
,
2984 goto bad_unshare_cleanup_cred
;
2986 if (new_fs
|| new_fd
|| do_sysvsem
|| new_cred
|| new_nsproxy
) {
2989 * CLONE_SYSVSEM is equivalent to sys_exit().
2993 if (unshare_flags
& CLONE_NEWIPC
) {
2994 /* Orphan segments in old ns (see sem above). */
2996 shm_init_task(current
);
3000 switch_task_namespaces(current
, new_nsproxy
);
3006 spin_lock(&fs
->lock
);
3007 current
->fs
= new_fs
;
3012 spin_unlock(&fs
->lock
);
3016 fd
= current
->files
;
3017 current
->files
= new_fd
;
3021 task_unlock(current
);
3024 /* Install the new user namespace */
3025 commit_creds(new_cred
);
3030 perf_event_namespaces(current
);
3032 bad_unshare_cleanup_cred
:
3035 bad_unshare_cleanup_fd
:
3037 put_files_struct(new_fd
);
3039 bad_unshare_cleanup_fs
:
3041 free_fs_struct(new_fs
);
3047 SYSCALL_DEFINE1(unshare
, unsigned long, unshare_flags
)
3049 return ksys_unshare(unshare_flags
);
3053 * Helper to unshare the files of the current task.
3054 * We don't want to expose copy_files internals to
3055 * the exec layer of the kernel.
3058 int unshare_files(struct files_struct
**displaced
)
3060 struct task_struct
*task
= current
;
3061 struct files_struct
*copy
= NULL
;
3064 error
= unshare_fd(CLONE_FILES
, ©
);
3065 if (error
|| !copy
) {
3069 *displaced
= task
->files
;
3076 int sysctl_max_threads(struct ctl_table
*table
, int write
,
3077 void __user
*buffer
, size_t *lenp
, loff_t
*ppos
)
3081 int threads
= max_threads
;
3083 int max
= MAX_THREADS
;
3090 ret
= proc_dointvec_minmax(&t
, write
, buffer
, lenp
, ppos
);
3094 max_threads
= threads
;